Institutional tax clienteles and payout policy

Institutional tax clienteles and payout policy

Journal of Financial Economics 100 (2011) 68–84 Contents lists available at ScienceDirect Journal of Financial Economics journal homepage: www.elsev...

415KB Sizes 0 Downloads 102 Views

Journal of Financial Economics 100 (2011) 68–84

Contents lists available at ScienceDirect

Journal of Financial Economics journal homepage: www.elsevier.com/locate/jfec

Institutional tax clienteles and payout policy$ Mihir A. Desai n, Li Jin Harvard University, Boston, MA 02163, USA

a r t i c l e in fo

abstract

Article history: Received 3 June 2008 Received in revised form 14 February 2009 Accepted 30 April 2009 Available online 12 November 2010

This paper employs heterogeneity in institutional shareholder tax characteristics to identify the relation between firm payout policy and tax incentives. Analysis of a panel of firms matched with the tax characteristics of the clients of their institutional shareholders indicates that ‘‘dividend-averse’’ institutions are significantly less likely to hold shares in firms with larger dividend payouts. This relation between the tax preferences of institutional shareholders and firm payout policy may reflect dividend-averse institutions gravitating towards low dividend paying firms or managers adapting their payout policies to the interests of their institutional shareholders. Evidence is provided that both effects are operative. Plausibly exogenous changes in payout policy result in shifting institutional ownership patterns. Similarly, exogenous changes in the tax cost of institutional investors receiving dividends results in changes in firm dividend policy. & 2010 Elsevier B.V. All rights reserved.

JEL classification: G20 G30 G35 H20 Keywords: Institutional investors Taxation Clienteles Payout policy

1. Introduction This paper employs heterogeneity in institutional shareholder tax characteristics to identify the relation between firm payout policy and institutional investor tax incentives. Institutional shareholders’ tax preferences vary, and firms, in turn, vary in the degree to which their investor bases are averse to dividends. By integrating dividend behavior with information on the institutional shareholder bases of firms, it is possible to analyze the degree to which the tax-based preferences of shareholders are associated

$ The authors thank seminar and conference participants at University of California at San Diego, Harvard University, Hong Kong Chinese University, New York University, 5th IDC Conference of Israel, 2008 Western Finance Association Meetings, 2008 NBER Summer Institute on Economics of Taxation and a number of colleagues, particularly James Poterba, for helpful comments on an earlier draft, and the Division of Research at Harvard Business School for financial support. n Corresponding author. E-mail address: [email protected] (M.A. Desai).

0304-405X/$ - see front matter & 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jfineco.2010.10.013

with firms’ dividend payment behavior. Emphasizing heterogeneity in tax preferences among institutional investors affords the opportunity to identify tax effects more cleanly relative to studies that employ the distinction between individuals and institutions and their presumed tax preferences. Across a variety of measures of the propensity of firms to pay dividends, the evidence indicates that there is a strong association between the composition of institutional shareholders by their tax preference and dividend payment behavior. This relation is robust to the inclusion of a variety of controls that have been shown to be associated with dividend paying behavior. Further analysis is performed to determine whether firms with a higher proportion of dividend-averse institutional investors are less likely to initiate dividends. Controlling for size, profitability, and investment opportunities, the share of dividend-averse institutional shareholders is negatively associated with the likelihood of initiating dividends. Such an association is consistent with two alternative explanations. First, institutional shareholders may be

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

sorting themselves across firms based on their dividend policies, i.e., tax-based dividend clienteles. Alternatively, the tax-based preferences of firm shareholders may be shaping the dividend policy of firms. To determine the degree to which ‘‘investor sorting’’ or ‘‘tax-based payout policy’’ is operative requires plausibly exogenous or predetermined variations in dividend policies or the tax preferences of shareholders. To investigate the sorting hypothesis, instrumental variables regressions are employed whereby a first-stage regression predicts dividend payments based on firm operating characteristics. These operating characteristics are plausibly unrelated to the degree to which taxable or non-taxable institutions are shareholders in the firms. Second-stage regressions that employ these predicted dividend payments suggest that the composition of institutional shareholders responds to exogenous changes in firm dividend policy. Institutional shareholders appear to sort on the basis of dividend policy. To investigate the relevance of the alternative hypothesis that firms are tailoring dividend policies to the preferences of their institutional shareholders, the analysis employs exogenous variation in the tax price of dividends relative to capital gains. This variation, created by legislated tax changes over the last two decades, is plausibly unrelated to preexisting dividend policies. Interactions of lagged changes in this tax price with the share of dividendaverse institutional shareholders are strongly associated with changes in dividend policy. Additional evidence also indicates that after a change in firm dividend payout policy, the concentration of dividendaverse institutions in the next year changes in response. Similarly, after the institutional investor base of a firm changes, firm dividend policy in the next year responds to the change in the dividend-averse institutional investor base. Overall, the empirical results reported in this paper suggest that causality runs in both directions: on the one hand, institutional shareholders appear to sort on the basis of firm dividend policy; on the other, corporate managers appear to cater to the preferences of their institutional shareholders. The remainder of the paper proceeds as follows. Section 2 provides a review of the extensive literature related to this subject. Section 3 describes the data, with particular emphasis on the classification of the institutions’ dividend preferences and the empirical specifications. Section 4 presents the results. Section 5 concludes. 2. Related literature This paper relates to two expansive literatures. The first examines the relation among clientele effects, institutional shareholders, and dividend policy, and the second explores the impact of the tax costs of dividends on firm payout policy. Beginning with Elton and Gruber (1970), a large literature examines market reactions to dividends to establish the role of dividend tax clientele.1 Such clientele effects 1

See Allen and Michaely (2003) for a review.

69

could arise from non-tax considerations, including informational advantages, distinct investment styles, or monitoring ability. Institutions may be better informed and this informational advantage may be manifested in differing attitudes toward payout policy. Amihud and Li’s (2006) study of the relation between price reactions to dividend announcements and institutional holding provides evidence that institutions are more informed. Michaely, Thaler, and Womack (1995) fail to find a significant change in institutional ownership after dividend omissions. Del Guercio (1996) examines the role of dividends in the portfolio selection of banks and mutual funds, and finds that dividend yield has no power in explaining the portfolio choice of these institutions. Her evidence suggests that the prudent-man rule has an important role, but that dividends do not play a major role in institutional investor portfolio decisions. On the other hand, Dhaliwal, Erickson, and Trezevant (1999) provide empirical evidence that after dividend initiation, the firms’ institutional investor clientele changes based on its tax preferences, with a surge in ownership by tax-exempt/tax-deferred and corporate investors. Hotchkiss and Lawrence (2007) report that institutions have distinct investment styles based on dividend yields. Grinstein and Michaely (GM) (2005) provide a comprehensive investigation of the relation between the concentration of institutional versus individual ownership and payout policy. GM consider a variety of factors that might affect payout policy, including institutional monitoring along the lines of Allen, Bernardo, and Welch (2000), free-cash flow problems (as in Jensen, 1986), taxation, regulatory changes, and adverse selection, in order to establish whether payout policy affects the willingness of institutions to invest in stocks, and whether a concentration of institutional holders in turn affects future payout policy. Although they find some evidence of the role of a variety of other factors, they do not find meaningful taxbased preferences between institutional investors and individual investors; that is, unlike some of the prior conjectures in the literature, there is no systematic evidence that individuals are averse to dividends because they are taxed more heavily. This evidence is consistent with the survey results in Brav, Graham, Harvey, and Michaely (2005) that show that institutional investors as a whole do not show a clear preference for dividends over repurchases. Relatedly, Jain (1999) reports that institutions prefer to invest in low-dividend-yield stocks, whereas individual investors prefer higher-dividend-yield stocks, inconsistent with the tax-based dividend clientele hypothesis that assumes institutions to be tax-advantaged. One reason that the studies broadly classifying investors into institutional investors versus individual investors fail to find significant evidence of a tax-related clientele effect could be that the classifications are too coarse. Indeed, Grinstein and Michaely (2005) acknowledge this possibility by pointing out that it is ‘‘possible that there is too much heterogeneity among institutions to capture this effect when we are looking at institutions as a whole or even at subgroups of institutions (such as pension funds).’’ Indeed, when focusing exclusively on the clientele effects of individual investors, Scholz (1992), Graham and Kumar

70

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

(2006), and Dahlquist, Robertsson, and Rydqvist (2006) are able to provide more direct evidence of the degree to which investors’ tax characteristics are associated with their portfolio holdings. This paper provides an institutional analogy to the above papers by exploring the heterogeneity of tax preferences within the class of institutional investors to investigate the role of tax-based clienteles. By comparing different types of institutions rather than making a typical individual versus institution comparison, tax explanations can be isolated relative to a variety of other explanations, including differential monitoring capacities, irrationality, and investor sophistication. Institutional investors are also increasingly more important in the holding of stocks, particularly large cap stocks, in the United States, and thus are more likely to be pivotal. Indeed, as reported in the survey results of Graham and Harvey (2001), corporate managers do respond to the demands of their institutional investors. Focusing on institutions and the impact of taxes on payout policy also enables this paper to significantly sharpen the measures of tax preferences and payout policy. Depending on their tax preferences, institutions can be classified into three groups: pass-through institutions for taxable investors who prefer capital gains over dividends; pass-through institutions for tax-exempt investors who are neutral; and corporations who prefer dividends over capital gains, as dividends and capital gains are treated differentially.2 Using a new database containing more precise classifications of institutions on the tax front, more precise estimates of the tax preferences of a firm’s aggregate institutional investors may be obtained.3 Two earlier papers attempt to use variation in institutional customer bases to study dividend behavior. Strickland (1996) finds that taxable institutional owners are more likely to own low-dividend-yield stocks, but also reports that tax-exempt investors do not appear to show a preference for either high- or low-yield stocks. Hotchkiss and Lawrence (2007) find that institutional investors seem to have distinct ‘‘dividend clienteles,’’ in that some institutions persistently hold stocks with high dividends. When firms announce dividend policy changes, dividend increases (decreases) are associated with increased (decreased) holdings by institutions that appear to prefer dividends based on their prior portfolio. They provide preliminary evidence that such dividend clientele seem to be at least partly related to tax considerations: for a small

2 See Desai and Gentry (2004) for a discussion of corporate responsiveness to capital gain treatment, and Barclay, Holderness, and Sheehan (2006) for an investigation of corporate preferences for dividends. 3 Several existing studies look into a more coarse classification of institutional investors, based on five reported types of institutional investors: banks, insurance companies, mutual funds, independent investment advisors, and others. The independent investment advisors account for a large portion of all the institutions. For example, in December 1997, the last year for which valid classifications of institutional investor types are available from the data provider Thomson Financial, there are a total of 1,489 institutions, of which 1,049 are listed as independent investment advisors. The more refined classification used in this paper allows better classification of the ‘‘independent investment advisors’’ and ‘‘others’’ categories.

subsample of investment managers that they can identify as serving taxable clients, they find some evidence that these institutions are less likely to increase their holdings around a dividend increase. Although tax-based dividend clientele and corporate payout policy are not central topics of their paper, Hotchkiss and Lawrence provide some evidence that is consistent with a tax-based dividend clientele hypothesis. This paper also contributes to the sizable literature that examines whether dividend tax changes influence dividend policy. Poterba (1987, 2004) employs time-series variations and economy-wide dividend payments, whereas more recent studies – including Blouin, Raedy, and Shackelford (2003), Brown, Liang, and Weisbenner (2007), and Chetty and Saez (2005) – examine how firms responded to the 2003 dividend tax cuts. These studies typically employ managerial shareholdings and institutional shareholdings to sharpen their hypothesis tests of the effects of the recent dividend tax cut. Bernheim and Wantz (1995) employ the time-series variation emphasized by Poterba to test the motivations for paying dividends, and Perez-Gonzalez (2003) uses the same variation to investigate whether large individual shareholders shape the dividend policies of the firms they own. To our knowledge, this paper is the first to employ variations in tax preferences across institutional shareholders to isolate the degree to which dividend policy at the firm level is related to the shareholders’ tax preferences over a longer period. It is also the first paper to seriously consider causality in both directions: whether corporate payout policy changes due to the tax costs of their investors, and whether investors self-sort into different tax-related dividend clienteles. 3. Data and methodology To analyze the relation between the tax preferences of institutional shareholders and dividend policy, detailed data on the characteristics of institutional shareholders are integrated with firm data on payout policy and operating characteristics. 3.1. Identifying the tax preferences of a firm’s institutional shareholders Whereas previous work has measured the tax preferences of shareholders as a function of whether they are largely institutions or retail investors, studies of the 2003 tax cut classify institutions somewhat more finely, but without reference to the tax preferences of the clients that these institutions serve. To determine the degree to which firms are held by dividend-averse investors, data on institutional investor holdings are combined with data on the tax preferences of the clients they serve. The institutional shareholder data used in this study are from the CDA/Spectrum 13F institutional investor holdings database. Investment managers who exercise investment discretion over $100 million or more in Section 13F securities must report to the Securities and Exchange Commission (SEC) on Form 13F for holdings of more than 10,000 shares or investments valued in excess of

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

$200,000.4 Data are available from 1980 to 2006, but because a number of institutions are improperly classified in 1998 and beyond, this paper limits the analysis to the 1980–1997 period.5 To make the analysis based on institutional investors meaningful, unless otherwise specified, the sample is further limited to firms for which at least 10% of the outstanding common shares are held by institutional investors identifiable in the 13F data. Spectrum reports five types of institutional investors: (i) bank trust departments, the clients of which can be either taxable or tax-exempt; (ii) insurance companies, which are taxable overall, but can also have tax-exempt accounts for clients; (iii) investment companies (open-end or closedend mutual funds), the clients of which can be either taxable or tax-exempt; (iv) independent investment advisors, whose clients can be either tax sensitive or non-tax sensitive; and (v) ‘‘others,’’ such as foundations, Employee Stock Ownership Plans (ESOPs), university endowments, and internally managed private and public pension funds, many of which are tax-exempt. For the purposes of this study, banks and investment companies (mutual funds) are classified as neither dividend-averse nor as non-dividend-averse, because no concrete information about their taxable investor bases can be obtained.6 Instead, these two categories are assumed to have the same ratio of dividend-averse shareholders imputed to the remaining institutional shareholders. Insurance companies are uniformly treated as being non-dividend-averse as they are in a corporate form and thus enjoy dividend reductions.7 Finally, investment advisors and the ‘‘other’’ category are classified on the basis of their client

4 Gompers and Metrick (2001) provide a detailed explanation of the 13F data. 5 As explained by Wharton Research Data Service, ‘‘The TYPECODE variable (which identifies institutions such as banks, insurance companies, investment companies, and independent investment advisors) is not reliable from 1998 forward. The reason is a mapping error that occurred when TFN integrated data from the former Technimetrics. The mapping error results into many institutions improperly classified as type 5 (endowments and ‘‘others’’). For example, in the last quarter of 1998 (DEC1998), the number of investment companies (typecode= 3) drops from 55,675 to 0 while typecode = 5 ’other institutions’ triples its constituency. These errors are carried forward. Additionally, the third quarter of 2007 (SEP2007) only reports 0.4% of the total sample. [The data vendor] regrets that these problems occurred but they have no plans to fix the problem.’’ As a robustness check, extrapolating the data for 1998 and beyond using the end of 1997 Spectrum classifications was also tried, creating a longer sample (from 1981 to 2006). The regression results from this expanded (but less accurate) data yield qualitatively similar results that are available upon request. 6 Another important consideration is that banks and mutual fund families typically consist of many separate accounts or funds. To the extent that these might lack the same level of tax sophistication, and that 13F reports only aggregate holdings data, documenting the existence of tax-sensitive trading using aggregate data is expected to be more difficult. The multiple account problems might exist in other types of institutions as well, but would be much less severe due to the nature of these businesses. 7 There are some caveats. Life insurance and property insurance companies, because their tax treatments are different, may have different reactions to capital gains tax overhang. Moreover, insurance companies invest in their own accounts large amounts of money for which they pay taxes, but they also invest on behalf of their clients through potentially tax-advantaged accounts. The 13F data do not differentiate between these. The tests reported in this paper are robust in terms of assuming that insurance companies have the same tax characteristics as mutual funds.

71

characteristics. For this purpose, institutional holdings data are supplemented with Investment Adviser Public Disclosure (IAPD) data obtained from the SEC. IAPD data contain the self-reported client bases of investment advisors broken down into ten categories: individuals (other than high net worth individuals); high net worth individuals; banking or thrift institutions; investment companies (including mutual funds); pension and profit-sharing plans (other than plan participants); other pooled investment vehicles (mostly hedge funds); charitable organizations; corporations or other businesses not listed above; state or municipal government entities; and ‘‘others’’ such as nonU.S. government entities. Investment advisors are required to report the percentage of business represented by each client category. Independent investment advisor preferences for dividends versus capital gains may differ depending on the nature of the clients they serve. An investment advisor is classified as dividend-averse if (1) there exist tax-sensitive investors capable of exerting pressure on managers or (2) the managers of the investment advisor might for personal reasons care about the tax consequences of the portfolios they manage. Among institutional investors who a priori might be expected to prefer capital gains over dividends are investment managers for high net worth individuals and hedge funds. The interests of the former are served by their relatively small number of clients, which facilitates communication and collaboration with and ultimately monitoring and the collective discipline of the fund managers. For incentive reasons, managers of hedge funds are typically required to invest a significant portion of their personal wealth in the funds they manage, and so for personal reasons, they are likely to care about the tax consequences of dividends versus capital gains.8 The information from IAPD is used to distinguish independent investment advisors who serve primarily dividend-averse clients from those who serve primarily non-dividend-averse clients. If an independent investment advisor has a clientele that consists primarily (Z50%) of high net worth individuals and hedge funds, it is classified as dividend-averse.9 Institutions are classified as taxexempt and thus, non-dividend-averse, if pension funds, state and local governments, and charitable organizations account for more than 50% of their clientele. This classification of dividend-averse and non-dividend-averse institutions

8 Many industry experts with whom we talked indicated that taxes are an important consideration for institutions. For example, Morgan White, managing director of Woodside Asset Management, Inc. which manages money for high net worth individuals and families, stated in an e-mail correspondence dated November 11, 2003: ‘‘Indeed, we do pay close attention to taxes when making selling decisions.’’ He went on to explain that they strictly follow ‘‘Highest-In, First-Out’’ selling of multiple lots to minimize tax liability, postpone realization of gains, and expedite realization of losses to offset gains realized in the same year. 9 Investment advisors who do not file the IAPD forms but who are identified in the hedge funds section of the quarterly publication Money Manager Directory are also included in this study. As reported by Brunnermeier and Nagel (2004), only institutions that conduct nonhedge-fund businesses, such as advising mutual funds and pension plans, are expected to file for the IAPD disclosure. That they are classified as hedge funds in the Money Manager Directory and not listed on the IAPD form indicates that the majority of their business is in hedge funds.

72

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

Share of dividend-averse institutional investors

0.50 0.40 0.30 0.20

by measuring the payout policy using the ratio of dividend payout as a proportion of total firm payout from both dividend and stock repurchase. Doing so yields qualitatively similar results.11 The detailed procedures for constructing the dividend-payout ratio, along with the rationale for using it to conduct the robustness checks, are contained in Appendix A.

0.10 Mean Median 0.00 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998

Fig. 1. The proportional of dividend-averse institutional investors, 1981–1997. This figure plots the share of dividend-averse institutional investors. The sample consists of all firms invested by institutional investors included in the CDA/Spectrum 13F institutional investor holding database, for the period of 1980–1997.

is fairly conservative. If clients cannot be clearly identified as either dividend-averse or non-dividend-averse, these investment advisors are not classified. For institutions categorized in 13F data as ‘‘other,’’ word searches are conducted for ‘‘pension,’’ ‘‘endowment,’’ ‘‘foundation,’’ and variations thereof, and any observation for which these words come up in their names is identified as non-dividendaverse. Finally, for each stock in each period, the proportion of dividend-averse investors is calculated as the ratio of identifiable dividend-averse institutional holdings to the sum of identifiable dividend-averse and non-dividendaverse institutional holdings.10 Fig. 1 shows the mean and median of the proportion of dividend-averse institutions among all identifiable institutions for the 1981–1997 period. There is considerable timeseries variation of both the mean and the median of the dividend-averse-institution proportion, and at any time, a substantial proportion of a typical firm’s institutional investors are dividend-averse: during this period the mean and the median are both within the range of 25–50% and there does not appear to be a significant time trend to the series. 3.2. Measuring payout policy For the main results reported in this paper, the payout policy of firms is measured using dividend yield so as to be consistent with the earlier literature. Dividend yield is defined as the total dollar amount of dividends declared on the common stock (Compustat industrial annual data item #21) divided by the market value of the common stock at year-end (Compustat industrial annual data item #24 times item #25). To deal with potential outliers in the dividend-yield measure due to very low stock prices, the measure is further winsorized at the 1% and 99% levels. Following Grinstein and Michaely (2005), the results are replicated using dividend yield defined as dividend divided by book assets. This does not qualitatively change the paper’s conclusion. Robustness checks are also performed 10 This calculation implicitly assumes that the unidentified portion of the institutional investors has the same proportional distribution as the identified portion. This assumption clearly is not precise. It might add noise, but not systematic bias, to the empirical analysis.

3.3. Empirical methodology The measures of the dividend-aversion of the institutional shareholders and the dividend yield are supplemented with additional explanatory variables. Following the literature, measures of cash holdings, return on assets, market capitalization, earnings, ratio of capital expenditures to assets, and leverage are all employed, as defined in Appendix B. In the main results reported in the paper, stock return volatility is also employed, calculated based on the daily stock returns during the previous fiscal year. As a robustness check, following Jagannathan, Stephens, and Weisbach (2000), ROA volatility, defined as the standard deviation of the ratio of operating income to total assets measured over the five-year period from years—4 through zero for any given year, is also employed.12 The results are not sensitive to the choice of volatility measure. Summary statistics for the variables described above are provided in Table 1. Prior to investigating the relation between institutional shareholder dividend- aversion and payout policy in a regression framework, it is useful to look for a relation in the raw data. Fig. 2 depicts such an effort. For each year, the sample of firms is partitioned into four quartiles according to the proportion of their identifiable institutional investors that are dividend-averse. Observations are then aggregated across years within these quartiles and a mean dividend yield is calculated for each quartile. By construction, there are an equal number of firms in each quartile. The figure plots the mean dividend yield against the quartiles’ average (mean) level of the proportion of dividend-averse institutional investors. Fig. 2 shows a negative relation between the proportion of dividendaverse institutions in a firm and the dividend yields of the firm in the raw data. The highest dividend-averse institutional investor concentration corresponds to an average dividend yield of 1.63%, the lowest corresponds to an average dividend yield of 2.27%. Table 2 addresses the concern that the dividend-averse institutional shareholder proxy might pick up risks or other characteristics of the firm. For example, one can imagine that tax-sensitive investors such as wealthy individuals might be less risk-averse than many tax-exempt institutions such as foundations. If that is the case, and if dividendpaying firms are less risky than non-dividend-paying firms, 11 This may not be entirely surprising, given that stocks with a high dividend yield tend to also have a high dividend payout ratio. In our sample, the empirical correlation between the dividend yield and the dividend payout ratio is 0.45, significant at the 0.1% level. 12 The drawback of the ROA volatility measure is that since not all firms have 5 years of ROA data, there potentially could be a survivorship bias.

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

73

Table 1 Summary statistics. Repurchase is the expenditure on the purchase of common and preferred stocks (Compustat Item #115) minus reduction in the value of the preferred shares outstanding (Item #56). Common dividends is the dollar dividends declared during the fiscal year (Item #21). Dividend yield is dividend divided by market capitalization, where market capitalization is Item #24 n Item #25. Dividend-payout ratio is dividend/(dividend +repurchase). Return on assets (ROA) is operating income (Item #13)/book value of total assets (Item #6). ROA standard deviation is the standard deviation of the ratio of operating income to total assets measured over the five-year period from year  4 to 0. Earnings is defined as earnings after interest and tax (Item #18). Ratio of capital expenditures to assets is the capital expenditures (Item #128) divided by total assets (Item #6). Market-to-book is book value of total assets (Item #6) plus market capitalization minus book value of equity (Item #216)), divided by book value of total assets. Ratio of debt to assets is (book value of total assets (Item #6)—book value of equity (Item #216))/book value of total assets (Item #6). Ratio of cash to assets is the cash and short-term investments (Item #1) divided by total assets. Ln(Sales) is the natural log of sales (Item #12). Variable

Number of observations

Mean

Standard deviation

1st Percentile

113,345 76,642 113,345

0.02 0.68 0.37

0.08 0.40 0.33

– – –

– 0.32 0.09

113,345 0.13 113,345 0.05 113,345 71.21 113,345 1,378.20 113,345 0.07 113,345 1.23 113,345 0.19 112,314 0.12 111,982 5.71

0.17 0.08 322.14 5,634.85 0.07 1.98 0.18 0.16 1.80

(0.38) 0.00 (140.00) 5.24 – 0.07 – 0.00 1.32

0.09 0.02 1.68 67.59 0.03 0.46 0.04 0.02 4.56

Dividend yield Dividend-payout ratio Share of dividend-averse institutional shareholders Return on assets (ROA) ROA standard deviation Earnings Market capitalization Ratio of capital expenditures to assets Market-to-book ratio Ratio of debt to assets Ratio of cash to assets Ln (Sales)

25th Median Percentile

75th Percentile

99th Percentile

0.01 0.91 0.27

0.03 1.00 0.61

0.09 1.00 1.00

0.14 0.03 10.06 208.46 0.06 0.80 0.16 0.06 5.64

0.19 0.06 40.87 780.40 0.09 1.42 0.28 0.17 6.88

0.39 0.33 1,265.98 20,932.30 0.33 6.77 0.75 0.75 9.91

Dividend yield

3.0% 2.5% 2.0% 1.5% 1.0% 1 Average share within quartile:

2.52%

2

3

17.81%

42.25%

4 87.02%

Quartiles based on share of dividend-averse institutional shareholders Fig. 2. Dividend across quartiles of dividend-averse institutional shareholding ratios. The bars present the average dividend yield across quartiles sorted on the basis of the share of dividend-averse institutional shareholders. The average share of dividend-averse institutional shareholders is presented below each quartile number. The quartiles are sorted within a year and then aggregated across years. The sample consists of all firms invested by institutional investors included in the CDA/Spectrum 13F institutional investor holding database, for the period of 1980–1997.

then it is possible that any negative correlation documented later between the share of dividend-averse institutional shareholders and dividend yield may be driven by differences in risk-aversion, not by differences in tax preferences. To determine whether the measure of dividend-averse institutional shareholder is correlated with other firm characteristics, for each year the sample is partitioned into four quartiles according to the share of dividend-averse institutional shareholders. For each quartile, the mean and the median are calculated for systematic risk (beta), return on assets (ROA), ROA standard deviation, earnings, market capitalization, ratio of capital expenditures to assets, market-to-book ratio, ratio of debt to assets, ratio of cash to assets, and log (Sales). Other than the measure related to payout policy, the stocks with higher concentrations of dividend-averse institutions and the stocks with lower concentrations of dividend-averse institutions do not differ in any systematic way in terms of other characteristics.

4. Regression results 4.1. Payout policy and institutional shareholder dividendaversion The specification in column 1 of Table 3 begins the analysis of the relationship between shareholder bases and dividend-payout behavior. The negative and highly significant coefficient indicates that changing the share of dividend-averse institutional shareholders from zero to one would be associated with a reduction of 0.4% in the dividend yield. The inclusion of additional controls in column 2 – ROA, firm performance volatility, earnings, market value, ratio of capital expenditures to assets, market-to-book ratio, and leverage – does not alter this basic finding. The specifications in columns 3 and 4 of Table 3 add industry controls; the association between institutional shareholdings and dividend yield persists.

74

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

Table 2 Breakdown of the firm characteristics by quartiles based on shares of dividend-averse institutional holdings. Beta is as calculated based on a market model regression over the previous one year using daily stock returns and the Center for Research in Security Prices (CRSP) value-weighted index as the market. Volatility of stock return is calculated as the standard deviation of daily stock returns over a year, annualized by multiplying with the square root of the number of trading days in the year. Return on assets is the ratio of operating income to total assets. ROA standard deviation is the standard deviation of ROA for the 5 years prior to the firm-year observation. Earnings is defined as earnings after interest and tax. Market capitalization is the market value of common shares outstanding. Ratio of capital expenditures to assets is the ratio of capital expenditures to total assets. Market-to-book ratio is the ratio of market value of assets to book value of assets. Ratio of debt to assets is the ratio of book debt to book assets. Ln (Sales) is the natural log of sales. Quartiles based on shares of dividend-averse institutional holdings Quartile 1 Variable Beta Volatility of stock return Return on assets (ROA) ROA standard deviation Earnings Market capitalization Ratio of capital expenditures to assets Market-to-book ratio Ratio of debt to assets Ratio of cash to assets Ln (Sales)

Quartile 2

Quartile 3

Quartile 4

Quartile 1

Mean value

Quartile 2

Quartile 3

Quartile 4

Median value

0.92 0.50 0.10 0.06 53.34

0.98 0.37 0.15 0.04 85.12

0.93 0.39 0.15 0.05 47.72

0.94 0.49 0.12 0.06 38.65

0.89 0.43 0.12 0.04 14.82

0.93 0.32 0.15 0.03 35.78

0.91 0.35 0.15 0.03 18.56

0.90 0.44 0.13 0.04 23.98

1187.03 0.07 1.16 0.19 0.12

1238.22 0.08 1.26 0.18 0.11

907.58 0.08 1.35 0.19 0.13

1110.20 0.07 1.15 0.19 0.13

120.23 0.05 0.73 0.16 0.06

286.21 0.06 0.89 0.16 0.05

350.57 0.06 0.89 0.16 0.06

168.29 0.05 0.72 0.15 0.06

5.36

6.75

6.00

4.74

5.24

6.75

6.02

4.77

To be consistent with the large body of literature in finance, the Fama and MacBeth (1973) methodology is also employed to compute robust standard errors on the coefficient estimates. The Fama-MacBeth methodology is a convenient and conservative way to account for potential cross-correlations in residuals. As reported by Fama and French (2002), the Fama-MacBeth standard errors potentially can be two to five times the Ordinary Least Square (OLS) standard errors from pooled panel data regressions that ignore residual cross-correlations. Cross-sectional regressions are first run for each year separately, controlling for the fixed effects at the industry level, and the time-series averages of the coefficient are used to draw inferences, using the time-series standard errors of the average slopes. Because the Fama-MacBeth procedure does not take into account residual autocorrelations, it is further corrected by following the procedure in Pontiff (1996) to adjust the time-series standard deviations to accommodate a first-order autocorrelation in the coefficient estimates. Columns 5–8 of Table 3 report the results using this Fama-MacBeth procedure, which confirm the results of the Tobit regression specifications. To assess the robustness of this association between dividend-averse institutional shareholders and payout policy, further investigations are performed to determine whether different cutoffs for the level of institutional investor holdings and the proportion of institutional holdings that are identifiable change the basic results. To ensure that institutional investors are the dominant shareholders in a firm, the specifications in Table 3 are preserved, but an additional requirement is imposed which requires that at least 50% (rather than 10%) of the outstanding shares be held by institutional investors. To ensure that the dividendaverse institutional classification does not over represent the true distribution of institutional investor types, the tests are repeated, requiring that at least 25% of the institutional holdings be identifiable (previously there

was no cutoff) as either dividend-averse or non-dividendaverse institutional holdings. Table 4 reports the results of the robustness checks with the requirement that there be both 50% institutional ownership of firms, and that 25% of these institutional holdings be identifiable. These results, on this significantly reduced sample size, confirm the patterns observed in Table 3. It is interesting to note that the coefficient estimates are almost three times higher than the corresponding specifications in Table 3. The coefficient estimates in column 4 of Table 4 indicate that changing the share of dividend-averse institutional shareholders from zero to one is associated with a decreased dividend yield of 1.3%. Given that the measurement issues are more limited and that the institutions are more likely to be pivotal in this sample, these results are reassuringly significantly larger than those in Table 3. 4.2. Does the share of dividend-averse institutional shareholders proxy for the tax preferences of institutional investors? To show that the share of dividend-averse institutional shareholders captures the cross-sectional variation in tax preferences, rather than other factors such as the riskaversion across institutional investors, the sample is split into periods of high-and-low dividend taxation and examines the correlation between the share of dividend-averse institutional shareholders and the dividend yield for each subsample. If this correlation is more negative during periods of high-dividend taxation, then it would support the tax clientele story. The rationale behind this test is that if taxes really matter, then investors’ aversion toward dividends should increase with the tax costs of dividends relative to capital gains. The sample is divided into two equal-length subsamples, based on the after-tax value of a dollar of dividends, relative to the after-tax value of a dollar of capital gains.

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

75

Table 3 The Relationship between dividend payout ratios and the proportion of dividend averse institutional investors. The dependent variable in these regressions is the dividend yield. Columns 1 through 4 are Tobit regressions with t-Statistcs based on standard errors clustered at the industry level. Columns 5–8 are Fama and MacBeth (1973) style regressions, with the resulting standard deviations further adjusted for potential autocorrelation in the time-series coefficient estimate, by adopting the method in Pontiff (1996) using AR(1) error terms. t-Statistics are below the coefficient estimates. ROA is the ratio of operating income to total assets. Firm-performance volatility is measured by the volatility of daily stock returns during the previous fiscal year. Earnings is defined as earnings after interest and tax. Market capitalization is the market value of common shares outstanding. Ratio of capital expenditures to assets is the ratio of capital expenditures to total assets. Market-to-book ratio is the ratio of market value of assets to book value of assets. Ratio of debt to assets is the ratio of book debt to book assets. For Fama-MacBeth regressions, we report the average adjusted R2. Dependent variable: Dividend yield (1)

(2)

(3)

(4)

(5)

Tobit regression Intercept Share of dividend-averse institutional shareholders

Fama-MacBeth regression

0.02 (4.49)  0.004 (5.39)

0.01 (3.59)  0.003 (4.69) 0.02 (10.70)  0.01 (4.09) 0.00 (6.25) 0.00 (1.37)  0.03 (6.65) 0.00 (9.23) 0.02 (14.26)

0.02 (3.18)  0.003 (3.25)

0.02 (3.26)  0.002 (2.59) 0.03 (4.68)  0.03 (3.27) 0.00 (2.60) 0.00 (1.58)  0.06 (6.11) 0.00 (1.59) 0.03 (1.41)

0.03 (3.53)  0.003 (3.44)

0.01 (6.87)  0.002 (2.28) 0.03 (9.38)  0.02 (6.02) 0.00 (6.14) 0.00 (1.89)  0.04 (11.58) 0.00 (5.57) 0.02 (2.72)

Y N

Y N

Y Y

Y Y

N N

N N

N Y

N Y

Earnings Market capitalization Ratio of capital expenditures to assets Market-to-book ratio Ratio of debt to assets

Number of observations

(8)

0.03 (4.06)  0.002 (3.03) 0.01 (8.67)  0.02 (6.27) 0.00 (7.43) 0.00 (2.94)  0.05 (12.74) 0.00 (3.77) 0.03 (22.71)

Firm-performance volatility

Pseudo R2

(7)

0.02 (5.16)  0.004 (5.54)

Return on assets (ROA)

Year fixed effects? Industry fixed effects?

(6)

0.003

0.016

0.021

0.031

0.003

0.012

0.018

0.026

113,345

113,345

113,345

113,345

113,345

113,345

113,345

113,345

Specifically, the paper follows Poterba (1987, 2004), who defines the tax-preference parameter (y) as the after-tax value of a dollar of dividends, relative to the after-tax value of a dollar of capital gains. If tdiv and tcg denote the marginal tax rates on dividends and long-term capital gains, respectively, then the dividend tax-preference parameter is given as yt ¼ ð1tdiv,t Þ=ð1tcg,t Þ. Analysis is performed on both subsamples. Consistent with the hypothesis that the tax preferences of institutions are the underlying force of the observed cross-sectional variations in firm payout policy, the correlation between the share of dividend-averse institutional shareholders and the dividend yield for the subsample with a lower y (i.e., higher tax costs for dividends relative to capital gains) is more negative. More formally, regression analysis is performed on the two subsamples, controlling for a variety of factors that might affect the dividend yield. As reported in Table 5, the coefficient on dividend-averse institutional shareholders is more negative in the subsample with a below-median y, and the hypothesis that the coefficients for the two subsamples are the same is rejected at the 1% level across a variety of regression specifications. The next analysis examines how the share of dividendaverse institutional shareholders affects market reaction surrounding the announcement of dividend increases. If the tax clientele story holds, the market reaction to these events should be negatively correlated with the share of

dividend-averse institutional shareholders because the value of the additional dividends should decline with the level of dividend-aversion. Following Grullon, Michaely, and Swaminathan (2002), cumulative abnormal returns (CAR) for the three days surrounding the dividend increase announcement are regressed on the share of dividendaverse institutional shareholders, the firm’s dividend yield, the magnitude of the dividend change, and the other firm characteristics used in the previous regression analysis. As shown in Table 6, controlling for other factors, the market reaction to these events is indeed more negative if the share of dividend-averse institutional shareholders is higher.

4.3. Do dividend changes drive changes in ownership patterns? The negative association between the concentration of dividend-averse institutions and the dividend yield provides direct evidence of clientele effects among institutions. This evidence does not, however, provide a precise answer as to whether institutional shareholders are sorting or whether the tax-based preferences of shareholders are dictating payout policy. To determine whether institutional shareholders are sorting across firms, Tables 7 and 8 illustrate how plausibly exogenous changes in dividend policy shift shareholding patterns.

76

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

Table 4 Robustness of relationship between dividend-aversion of institutional shareholders and payout policy to different inclusion criteria. The dependent variable in these regressions is the dividend yield. Columns 1–4 are Tobit regressions with t-statistics based on standard errors clustered at the industry level. Columns 5–8 are Fama and MacBeth (1973) style regressions, with the resulting standard deviations further adjusted for potential autocorrelation in the time-series coefficient estimate, by adopting the method in Pontiff (1996) using AR(1) error terms. t-Statistics are below the coefficient estimates. ROA is the ratio of operating income to total assets. Firm-performance volatility is measured by the volatility of daily stock returns during the previous fiscal year. Earnings is defined as earnings after interest and tax. Market capitalization is the market value of common shares outstanding. Ratio of capital expenditures to assets is the ratio of capital expenditures to total assets. Market-to-book ratio is the ratio of market value of assets to book value of assets. Ratio of debt to assets is the ratio of book debt to book assets. For Fama-MacBeth regressions, we report the average adjusted R2. Dependent variable: Dividend yield (1)

(2)

(3)

(4)

(5)

(6)

Tobit regression Intercept Share of dividend-averse institutional shareholders

0.01 (1.23)  0.015 (6.44)

0.02 (1.72)  0.014 (5.94) 0.03 (4.02)  0.03 (2.13) 0.00 (1.63) 0.00 (0.60)  0.09 (7.85) 0.00 (6.14) 0.01 (1.93)

0.02 (3.65)  0.009 (5.03)

0.02 (2.32)  0.005 (3.86) 0.03 (1.70)  0.07 (3.15) 0.00 (1.14) 0.00 (1.09)  0.08 (2.95) 0.00 (0.42) 0.05 (0.71)

0.02 (8.53)  0.010 (4.59)

0.03 (9.22)  0.013 (5.40) 0.06 (4.56)  0.03 (2.36) 0.00 (0.83) 0.00 (1.81)  0.06 (6.72)  0.01 (1.80)  0.01 (2.79)

Y N

Y N

Y Y

Y Y

N N

N N

N Y

N Y

Earnings Market capitalization Ratio of capital expenditures to assets Market-to-book ratio Ratio of debt to assets

Number of observations

Fama-MacBeth regression

0.03 (3.06)  0.006 (3.61) 0.01 (1.88)  0.07 (4.15) 0.00 (1.58) 0.00 (1.14)  0.05 (5.27) 0.00 (2.70) 0.03 (7.86)

Firm-performance volatility

Pseudo R2

(8)

0.02 (2.35)  0.013 (5.85)

Return on assets (ROA)

Year fixed effects? Industry fFixed effects?

(7)

0.007

0.031

0.080

0.099

0.008

0.022

0.068

0.083

11,395

11,395

11,395

11,395

11,395

11,395

11,395

11,395

To test whether exogenous changes in dividend yields attract dividend-averse institutional investors, the proportion of dividend-averse institutional investors is regressed on the predicted dividend yields generated by a first-stage regression, using cash levels and firm-performance volatility as independent variables.13 The premise in these specifications is that the variables are associated with changes in payout policy, but are not directly related to the degree to which dividend-averse institutional shareholders are drawn to firms.14 It is important to note that this is a setting in which it is particularly useful to focus on the share of institutional shareholders that are

13 Brav, Graham, Harvey, and Michaely (2005) show that the cash level has a negative effect on dividends and a positive effect on repurchases. Jagannathan, Stephens, and Weisbach (2000) and Guay and Harford (2000) find that firms that pay dividends have more stable earnings than firms that use share repurchases. They therefore conclude that share repurchases are used to pay out extraordinary transitory earnings and dividends are used to pay out permanent earnings. 14 Although institutional holdings in the aggregate might be affected by the cash level or the volatility of firm performance, there is no evidence suggesting that the composition of institutions – dividend-averse versus non-dividend-averse – is directly affected by the cash level or by firmperformance volatility. Indeed, as we saw in Table 2, there is no systematic difference in these characteristics across firms with different concentrations of dividend-averse institutional shareholders.

dividend-averse rather than only the share of institutional shareholders per se. The results of the first-stage regressions are provided in Table 7, Panel A. When the dividend yield is regressed on the ratio of cash to assets and firm-performance volatility, the coefficient on cash is negative and statistically significant, indicating that firms with more cash pay out disproportionately less through dividends. The coefficient on firm-performance volatility is also negative, indicating that firms with more volatile performance pay out disproportionately less through dividends. This pattern is robust to the type of regression, whether univariate or multivariate. The predicted values from the regressions with both instruments (i.e., cash-asset ratio and firmperformance volatility) are employed in the second-stage regressions discussed below. Second-stage results are provided in Table 7, Panel B. The coefficients on the predicted values of the dividend yield are negative and highly significant. These results are fairly stable across specifications that include industry controls and additional controls for other determinants of institutional shareholding patterns. In the specification in column 2, the coefficient estimate indicates that a change of 10% in dividend yield would result in a 19% reduction in the proportion of institutional shareholders that are dividend-averse. These results are robust to the use

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

77

Table 5 The Comparison of the regression coefficients for the share of dividend-averse institutional shareholders, for the the high-tax cost and low-tax cost subsamples. This table summarizes the regression results following the specification in Table 3, but carried out for the two subsamples based on relative costs of dividends, measured using the theta obtained from Poterba (2004). The high-tax cost subsample includes years 1981, 1982, 1983, 1984, 1985, 1986, 1995, and 1997, and the low-tax cost subsample includes years 1987, 1988, 1989, 1990, 1991, 1992, 1993, 1994, and 1996. The dependent variable in these regressions is the ratio of dividends to the sum of dividends and repurchases. Columns 1–4 are Tobit regressions with t-statistics based on standard errors clustered at the industry level. Columns 5–8 are Fama and MacBeth (1973) style regressions, with the resulting standard deviations further adjusted for potential autocorrelation in the time-series coefficient estimate, by adopting the method in Pontiff (1996) using AR(1) error terms. t-Statistics are below the coefficient estimates. The coefficients for the two subsamples are significantly different at the 1% level across all regression specifications. Dependent variable: Dividend yield (1)

(2)

(3)

(4)

(5)

(6)

Tobit regression High tax cost subsample Low tax cost subsample Controls on other firm characteristics? Year fixed effects? Industry fixed effects?

(7)

(8)

Fama-MaBbeth regression

 0.006 (8.16)  0.003 (2.36)

 0.003 (6.18)  0.001 (1.83)

 0.005 (7.85)  0.002 (3.01)

 0.004 (7.01)  0.001 (3.12)

 0.004 (5.02)  0.002 (1.68)

 0.003 (4.91)  0.001 (1.83)

 0.004 (4.89)  0.002 (1.90)

 0.004 (5.49)  0.001 (1.71)

N Y N

Y Y N

N Y Y

Y Y Y

N N N

Y N N

N N Y

Y N Y

Table 6 Test of market reaction to dividend increase announcements. The dependent variable in these regressions is the 3-day cumulative abnormal return in percent with respect to the NYSE/Amex value-weighted index around the dividend announcement, defined according to Grullon, Michaely, and Swaminathan (2002). Columns 1 and 2 are OLS regressions with t-statistics based on standard errors clustered at the industry level. Columns 3 and 4 are Fama and MacBeth (1973) style regressions, with the resulting standard deviations further adjusted for potential autocorrelation in the time-series coefficient estimate, by adopting the method in Pontiff (1996) using AR(1) error terms. t-Statistics are below the coefficient estimates. Share of dividendaverse institutional shareholders is defined as in Table 1. The magnitude of dividend change is the percentage change in dividend. ROA is the ratio of operating income to total assets. Firm-performance volatility is measured by the volatility of daily stock returns during the previous fiscal year. Earnings is defined as earnings after interest and tax. Market capitalization is the market value of common shares outstanding. Ratio of capital expenditures to assets is the ratio of capital expenditures to total assets. Market-to-book is book value of total assets plus market capitalization minus book value of equity, divided by book value of total assets. Ratio of debt to assets is the ratio of book debt to book assets. For Fama-MacBeth regressions, we report the average adjusted R2. Dependent variable: 3-day CAR (1)

(2)

(3)

Tobit regression Share of dividend-averse institutional shareholders Dividend yield Magnitude of dividend change Return on assets (ROA) Firm-performance volatility Earnings Market capitalization Ratio of capital expenditures to assets Market-to-book ratio Ratio of debt to assets Year fixed effects? Industry fixed effects?

(4)

Fama-MacBeth regression

 0.005 (9.96) 0.07 (9.67) 0.002 (4.79)  0.01 (3.64) 0.01 (3.21) 0.00 (5.08) 0.00 (5.31)  0.01 (2.52) 0.00 (0.63) 0.00 (0.17)

 0.006 (10.27) 0.06 (7.87) 0.003 (5.18)  0.01 (4.25) 0.01 (2.90) 0.00 (0.58) 0.00 (0.91)  0.01 (2.09) 0.00 (1.12) 0.00 (0.82)

 0.006 (5.40) 0.11 (3.85) 0.002 (2.17)  0.01 (1.60)  0.01 (0.48) 0.00 (0.35) 0.00 (0.23) 0.00 (0.36) 0.00 (1.50) 0.00 (0.34)

 0.006 (9.38) 0.11 (7.68) 0.004 (3.68)  0.02 (4.31)  0.01 (1.01) 0.00 (0.07) 0.00 (1.35)  0.01 (1.32) 0.00 (3.33) 0.00 (0.01)

Y N

Y Y

N N

N Y

Pseudo R2

0.054

0.062

0.032

0.033

Number of observations

8,592

8,592

8,592

8,592

78

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

of the Fama-MacBeth procedure discussed above, as indicated by the results in columns 3 and 4. In short, changes in dividends that are plausibly unrelated to the composition of dividend-averse institutional investors are associated with changes in the institutional investor composition. Institutional investors appear to self-sort based on the tax consequences of the firm payout policy. 4.4. Do the preferences of institutional shareholders drive dividend policy? Although the results presented in Tables 7 and 8 indicate that institutional investors are sorting across

firms, it is still possible that, at the same time, firms also respond to the tax-based dividend preferences of their institutional shareholders through changes in payout policy. To determine if such a mechanism is operative, the next analysis investigates whether dividend policy responds to arguably exogenous changes in the tax preferences of firm institutional investors. These changes are identified as those arising from changes in the relative cost of dividends to capital gains and the proportion of firm institutional investors that are dividend-averse. To measure the time-series change in the investor preferences for dividends relative to capital gains, the analysis again relies on the tax-preference parameter (y)

Table 7 Does exogenous changes in dividend drive changes in ownership pattern? Panel A lists the first stage results for determinants of dividend payout policy. The dependent variable in these regressions is the dividend yield. Volatility of firm performance is the standard deviation of daily stock return during the fiscal years prior to the firm-year observation. Ratio of cash to assets is the ratio of cash to total assets. The regressions also control for all other firm characteristics used in Table 2. t-Statistics based on standard errors clustered at the industry level are below the coefficient estimates. Panel B lists the second stage results for effects of changes of dividend payout policy on Insitutional shareholder bases. The dependent variable in these regressions is the the ratio of identifiable dividend-averse institutional holdings to the sum of identifiable dividend-averse and non-dividend-averse institutional holdings. Columns 1–4 are OLS specifications with t-Statistcs based on standard errors clustered at the industry level. Columns 5–8 are Fama and MacBeth (1973) style regressions, with the resulting standard deviations further adjusted for potential autocorrelation in the time-series coefficient estimate, by adopting the method in Pontiff (1996) using AR(1) error terms. t-Statistics are below the coefficient estimates. ROA is the ratio of operating income to total assets. Firm-performance volatility is measured by the volatility of daily stock returns during the previous fiscal year. Earnings is defined as earnings after interest and tax. Market capitalization is the market value of common shares outstanding. Ratio of capital expenditures to assets is the ratio of capital expenditures to total assets. Market-to-book ratio is the ratio of market value of asset to book value of assets. Ratio of debt to assets is the ratio of book debt to book assets. For Fama-MacBeth regressions, we report the average adjusted R2. Dependent variable: Dividend yield (1)

(2)

Panel A: first-stage results for determinants of dividend payout policy Intercept 0.02 (77.25) Volatility of firm-performance Ratio of cash to assets

(3)

0.03 (72.36)  0.54 (38.68)

0.03 (74.35) -0.51 (35.02)  0.02 (12.36)

113,345

113,345

113,345

0.004

0.010

0.012

 0.03 (22.52)

Number of observations Pseudo R2

Dependent variable: Share of dividend averse institutional shareholders Tobit regression (1)

Fama-MacBeth regression (2)

(3)

Panel B: Second-stage results for effects of changes of dividend payout policy on insitutional shareholder bases Intercept  8.58  9.56 (20.56) (22.87) Dividend yield  1.77  1.95 (15.45) (8.51) Market capitalization 0.00 0.00 (37.31) (38.94) Ratio of capital expenditures to assets  0.05  0.03 (3.78) (2.51) Market-to-book ratio 0.00 0.00 (3.34) (5.38) Ratio of debt to assets  0.01 0.00 (2.43) (0.60) Year fixed effects? Industry fixed effects? Adjusted R2 Number of observations

(4)

 7.39 (12.35)  1.84 (12.54) 0.00 (30.32)  0.07 (3.32) 0.00 (3.13)  0.01 (2.12)

 8.51 (20.09)  1.93 (7.01) 0.00 (36.31)  0.04 (3.38) 0.00 (5.07) 0.00 (0.72)

Y N 0.016

Y Y 0.015

N N 0.015

N Y 0.017

113,345

113,345

113,345

113,345

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

79

Table 8 The effect of tax changes on dividend payout policy. The dependent variable is the change of the log of dividend yield from the previous year. The independent variables are change and lagged variables of the log of sales, the change and lagged variables of the log of the tax-preference parameter, the log of lagged overall tax-price parameter, namely, Poterba’s (2004) y, and the log of lagged ratio of dividend to total payout. The tax-preference parameter is defined as the weighted-average value of dividends relative to capital gain for institutional investors as measured by (ynproportion_dividend_averse_institution + 1nproportion_dividend_neutral_institution). t-Statistics based on robust standard errors are reported under each coefficient. Dependent variable: change of log of dividend yield Regression without industry fixed effects (1) Change in log of sales Change in log of weighted-average tax preference for dividend Log of lagged dividend yield Log of lagged sales Log of lagged investor tax preference

(2) 0.07 (2.95) 0.13 (1.05)  0.09 (31.53) 0.05 (11.59) 0.22 (2.70)

Log of lagged y Adjusted R2 Number of observations Estimated long-run elasticity of dividend-payout ratio to tax incentives

proposed by Poterba (1987, 2004). Poterba (2004) uses aggregate data to study how a time-series change in the weighted-average tax-preference parameter affects aggregate dividend payments. He estimates a model similar to Lintner’s (1956) partial adjustment model of corporate dividends. The annual change in real dividends, Dln Dt, is modeled as a function of the change in corporate profits (Dln Profitt) and the change in the relative tax burden on dividends versus capital gains (Dln yt). Changes in dividends are also a function of lagged levels of profits, dividends, and the relative tax burden. The specification Poterba employs is

Dln Dt ¼ b0 þ b1 ðDln Profitt Þ þ b2 ðDln yt Þ þ b3 ðln Dt1 Þ þ b4 ðln Profitt1 Þ þ b5 ðln yt1 Þ þ et : As explained in Poterba (2004), the estimated long-run elasticity of the dividend yield with respect to the weighted-average tax price is  b5/b3. Using time-series analysis of economy-wide dividend payments, he estimates the long-run elasticity of dividend yields with respect to the weighted-average tax price to be 3.9–5.1, depending on how precisely profit is accounted for. This suggests that a one percentage point increase in the taxpreference parameter will increase the dividend yield by 3.9% (or 5.1%) in the long run. A tax-preference parameter for corporate investors can be similarly defined. As Poterba (2004) and Barclay, Holderness, and Sheehan (2006) show, there is little evidence that corporate investors affect and/or react to firm dividend policy. One possible reason is that because corporate investors invest for strategic reasons, dividend policy is at best a second-order consideration. For simplicity, the corporate investor tax-preference parameter is excluded from the specification. In additional (unreported)

Regression with industry fixed effects (3)

0.07 (2.95) 0.14 (1.11)  0.09 (31.32) 0.05 (11.52) 0.18 (2.89)  0.17 (1.49)

(4)

0.07 (3.14) 0.04 (0.34)  0.10 (34.40) 0.06 (12.77) 0.18 (2.31)

0.07 (2.96) 0.14 (1.12)  0.09 (31.27) 0.05 (11.54) 0.18 (2.20)  0.16 (1.42)

0.15

0.13

0.16

0.16

21,772

21,772

21,772

21,772

2.56

2.14

1.72

2.16

tests that include these parameters, they are not significant. Furthermore, no other parameter’s significance level is materially affected by the inclusion of the corporate taxpreference parameter. Whereas Poterba’s (2004) evidence is at the macro level, this paper employs the measure of a firm’s institutional investor distribution (between dividend-averse and nondividend-averse investors) to further test whether at the firm level, the tax-preference change, combined with the change in the proportion of dividend-averse institutional investors, is likely affecting the firm’s dividend policy. Specifically, the measure of the proportion of the dividendaverse institutional investors is interacted with the Poterba tax-preference parameter to form a measure of overall ‘‘representative investor tax preference’’ as the weightedaverage after-tax value of dividends. In particular, the representative investor tax preference is defined as T¼y  proportion_dividend_averse + 1  (1 proportion_ dividend_averse). The intuition is that for the dividend-averse investors in the firm, the actual after-tax value of a dollar of dividends is only $y ( o$1); for the non-dividend-averse investors in the firm, the actual after-tax value of a dollar of dividends is $1. If a firm does indeed react to the tax consequences of its representative institutional investor, then as the ‘‘representative investor tax preference’’ increases, the dividend payment will increase. The regression specification used in the analysis is

Dln Dit ¼ b0 þ b1 ðDln salest Þ þ b2 ðDln Tit Þ þ b3 ðlnDt1 Þ þ b4ðln salest1 Þ þ b5 ðln Tt1 Þ þ b6 ðln yt1 Þ þ et , where Dit is the dividend yield for a firm-year observation. Prior to discussing the results, several points should be noted. In the regression specification, the household tax

80

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

parameter (yt) has been changed to the weighted-average tax-preference parameter (Ti,t) so as to generate both timeseries and cross-sectional variations in the tax environment of the firms’ representative institutional investors. The natural log of profit has been changed to natural log of sales revenue, as profit can be negative for individual firms (but not for the aggregate data that Poterba employs). The use of sales in place of (sometimes negative) profit is routine, as in Perez-Gonzalez (2003) and Grinstein and Michaely (2005). Because the paper exploits the timeseries variation of the relative price of dividends to capital gains (the Poterba theta), the Fama-MacBeth analysis can no longer be used; instead the year fixed effect is added, along with the industry fixed effect (because dividend policy is likely correlated with industry characteristics), and firm-level controls used in other studies such as PerezGonzalez (2003) and Grinstein and Michaely (2005). The specification is tried both with and without controls on market-to-book ratio, asset growth, and beta-adjusted excess return of the stock. The main results are not sensitive to these additional controls. As in the Poterba (2004) specification, the short-run effect of a tax-preference change on dividend yield is captured by b2, the longrun effect is captured by  b5/b3. The results for this specification are reported in Table 8. Columns 1 and 3 of Table 8 do not include the lagged value of y, and columns 3 and 4 add industry controls. The shortrun elasticity of dividend policy to tax-preference change is weak and statistically insignificant, corroborating Poterba’s findings: the coefficient estimate of b2 ranges from 0.04 to 0.14, but it is never statistically significant. In the longer run, there is a significant response to the taxpreference change of investors. Using a calculation similar to Poterba’s (2004), the long-run elasticity of dividend yield to the change in representative investor dividend valuation is 2.56 ( = 0.22/0.09) in the first column and 1.72 ( = 0.18/ 0.10) in the third column. These results hold even after directly controlling for the time-series variation of the tax-preference parameter (yt), as reported in columns 2 and 4.15 Firms respond to the tax preferences of their institutional investors, at least in the long run.

4.5. Robustness checks on the causality tests The previous tests provide evidence that causality can run both ways, namely, institutional shareholders self-sort into firms with different payout policies depending on their tax status (or the tax status of their clients), and firm payout policy also responds to their institutional shareholders’ tax preferences. As robustness checks, another set of tests is provided for the hypothesis that predetermined changes in firm payout policy lead to changes in institutional shareholder concentration, and that predetermined changes in the firms’ institutional investor dividend preference lead to changes in the firms’ payout policies. 15 Although the paper reports the long-run elasticity in the same fashion as in columns 1 and 3, the estimated long-run elasticity in columns 2 and 4 is less interpretable given that y is also directly controlled for.

A test is first run on the hypothesis that predetermined changes in firm payout policy lead to changes in institutional shareholder concentration. The predetermined changes in the firm dividend-payout policy are measured, and then an empirical analysis is conducted to determine whether these changes affect the changes in the concentration of dividend-averse institutional investors in the subsequent period. To make sure the analysis is picking up the causality in the right direction (changes in firm dividend-payout policy causing changes in the compensation of dividend-averse institutional shareholders), the change in the concentration of dividend-averse institutional investors during the current year (measured as the difference between the current level and the level 12 months earlier) is regressed on the change in the firm’s dividend yield during the previous year (measured as the difference between the average dividend yield during the previous four quarters and the average dividend yield during the four quarters before that). For example, the change in the dividend-averse institutional shareholder concentration from December 2003 to 2004 is regressed on the difference in the average dividend yield between the year 2003 (averaged over all four quarters) and the year 2002, after controlling for a variety of other factors. A negative and significant coefficient on the dividend yield will then suggest that, controlling for everything else, if firms increase their propensity to pay dividends, some dividend-averse institutions will shy away from those firms. The regression specification employed in Table 7, Panel B is employed, with the exception that the dependent variable, the proportion of dividend-averse institutional shareholders, is now replaced with the change in the share of dividend-averse institutional shareholders during the current year; similarly, instead of using the predicted level of the dividend yield through a first-stage instrumental variable regression, the change in the dividend yield during the previous year is now used. The regression results are summarized in Table 9. As demonstrated by Table 9, a predetermined change in the firm dividend policy is associated with a significant change in the concentration of dividend-averse institutional investors. For example, in a Fama-MacBeth regression setting, controlling for all other factors and controlling for industry fixed effects, the coefficient on the changes in average dividend yields 2 years before is  0.15 (column 8), suggesting that a change in the average dividend yield from zero to one in the past decreases the proportion of dividend-averse institutional investors in the future by 15%. A test is then run on the hypothesis that predetermined changes in the institutional shareholder concentration lead to changes in the firm payout policy. The predetermined changes in the concentration of dividend-averse institutional investors is measured first, and then analysis is carried out to see whether these changes affect the changes in the firm dividend-payout policy in the subsequent period. To make sure the causality in the right direction is picked up (changes in the composition of the investors causing changes in the firm dividend policy), the change in the firm’s dividend-payout policy in the current year is regressed on the change in the firm’s proportion of

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

81

Table 9 The relationship between change in proportion of dividend averse institutional investors and the changes in dividend yields. The dependent variable is the change during the past year in the proportion of dividend-averse institutional investors. The independent variable is the change in annual average dividend yield 2 years before. For example, we regress year 2003 changes on dividend-averse investor proportion on the changes in average dividend from 2001 to 2002. Columns 1–4 are OLS with t-Statistcs based on standard errors clustered at the industry level. Columns 5–8 are Fama and MacBeth (1973) style regressions, with the resulting standard deviations further adjusted for potential autocorrelation in the time-series coefficient estimate, by adopting the method in Pontiff (1996) using AR(1) error terms. t-Statistics are below the coefficient estimates. ROA is the ratio of operating income to total assets. Market capitalization is the market value of common shares outstanding. Ratio of capital expenditures to assets is the ratio of capital expenditures to total assets. Market-to-book ratio is the ratio of market value of assets to book value of assets. Ratio of debt to assets is the ratio of book debt to book assets. Dependent variable: Changes in proportion of dividend averse institutional investors during the past 12 months (1)

(2)

(3)

(4)

(5)

(6)

Tobit regression Intercept Changes in average dividend yield 2 years before

0.38 (27.65)  0.116 (3.04)

0.41 (29.35)  0.115 (3.80) 0.00 (5.53) 0.00 (12.16) 0.03 (1.42) 0.00 (0.78)  0.01 (1.01)

0.31 (32.36)  0.170 (3.61)

0.23 (27.35)  0.157 (3.15) 0.00 (1.29) 0.00 (6.53) 0.02 (0.34) 0.00 (1.24)  0.01 (0.76)

0.38 (42.42)  0.141 (3.19)

0.41 (40.20)  0.152 (2.96) 0.00 (3.88) 0.00 (14.86) 0.02 (1.61) 0.00 (4.76)  0.01 (2.34)

Y N

Y N

Y Y

Y Y

N N

N N

N Y

N Y

Ratio of capital expenditures to assets Market-to-book ratio Ratio of debt to assets

Number of observations

Fama-MacBeth regression

0.31 (16.94)  0.158 (5.41) 0.00 (4.74) 0.00 (19.61)  0.01 (0.38) 0.00 (3.09)  0.01 (1.23)

Market capitalization

Adjusted R2

(8)

0.35 (35.15)  0.128 (5.13)

Earnings

Year fixed effects? Industry fixed effects?

(7)

0.01

0.02

0.03

0.05

0.02

0.03

0.06

0.06

77,465

77,465

77,465

77,465

77,465

77,465

77,465

77,465

dividend-averse institutions in the year before that, controlling for a variety of other factors. A negative and significant coefficient on the proportion of dividend-averse institutions will then suggest causality in one direction: if institutions are self-selecting depending on their dividend preferences, it might be argued that firm dividend policy changes can still affect the contemporaneous composition of the firms’ dividend-averse institutional investors. However, it is hard to argue that institutions are acting on anticipated dividend changes when deciding on what firms to hold.16 To undertake this analysis, the specification in Table 3 is adopted, but instead of using as a dependent variable the current period level of the dividend yield, the change of the dividend yield from a year earlier is now used (namely, the changes in the average ratios from the last four quarters to the previous four quarters are measured). Similarly, instead of using as the independent variable the current period level of the proportion of dividend-averse institutions, the new analysis now uses the change in that proportion that occurred a year before the starting point of the measurement of the dividend-yield change. For example, the average dividend yield from 2002 to 2003 is regressed

16 As long as dividend-averse institutions can respond after the dividend policy changes are adopted, there is no apparent need for them to make a stock selection decision based on the anticipated dividend policy change of the firm, because such anticipations may be risky.

on the change in the proportion of dividend-averse institutional investors during 2001. The regression results are shown in Table 10. As might be expected, since Table 3 compounds the causality in both directions and Table 10 only captures the causality in one way, the magnitude and significance of the coefficients on the changes in the composition of dividend-averse institutions reported in Table 10 are smaller than those reported in Table 3. Nonetheless, the coefficients are significant both statistically and economically. Controlling for a variety of other factors, changes in the dividend yield this year respond negatively to the (predetermined) changes in the proportion of dividend-averse institutions in the previous year: an increase in the proportion of dividendaverse institutions in the previous year leads to a decrease in the firm’s dividend yield in the current year. This evidence is again consistent with the hypothesis that the dividend preferences of the institutional investor cause changes in the firms’ dividend-payout policy. 4.6. The relation between dividend clientele and the decision to pay dividends As a piece of corroborating evidence that tax-based dividend clienteles affect firm payout policy, evidence is also provided on the relation between dividend clienteles and the decision to pay dividends. This is done by a logit analysis of the decision to pay dividends as a function of the share of dividend-averse institutional shareholders and other firm

82

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

Table 10 The Relationship between change in dividend yields and the changes in proportion of dividend-averse institutional investors. The dependent variable is the change during the past year in annual average dividend yield. The independent var. is the change in the proportion of dividend-averse investors 2 years before. For example, we regress year 2003 changes in average dividend payout ratio on changes on dividend-averse investor proportion from 2001 to 2002. Columns 1–4 are OLS with t-Statistcs based on standard errors clustered at the industry level. Columns 5–8 are Fama and MacBeth (1973) style regressions, with resulting standard deviations further adjusted for potential autocorrelation in time-series coefficient estimate, by adopting the method in Pontiff (1996) using AR(1) error terms. t-Statistics are below the coefficient estimates. ROA is the ratio of operating income to total assets. Firm-performance volatility is measured by the volatility of daily stock returns during the previous fiscal year. Earnings is earnings after interest and tax. Market capitalization is the market value of common shares outstanding. Ratio of capital expenditures to assets is the ratio of capital expenditures to total assets. Market-to-book ratio is the ratio of market value of assets to book value of assets. Ratio of debt to assets is the ratio of book debt to book assets. Dependent Variable: Changes in Dividend Yield during the past 12 months (1)

(2)

(3)

(4)

(5)

Tobit Regression Intercept Changes in share of dividend-averse institutional shareholders 2 years before Return on assets (ROA) Firm-performance volatility Earnings Market capitalization Ratio of capital expenditures to assets Market-to-book ratio Ratio of debt to assets Year fixed effects? Industry fixed effects? Adjusted R2 Number of observations

characteristics. The exact procedure outlined in Fama and French (2001) is followed, particularly Table 5 in their paper and the variable definitions therein. The variable on the proportion of dividend-averse institutions is added for the analysis. The annual logit regressions show the marginal effects of the share of dividend-averse institutional shareholders, size, profitability, and investment opportunities on the likelihood that a firm pays dividends. As in Fama and French (2001), the size of an NYSE, Amex, or Nasdaq firm for a given year is its NYSE percentile, NYPt, that is, the percentage of NYSE firms that have the same or smaller market capitalization. This size measure is meant to neutralize any effects of growth in typical firm size over time. Profitability is the ratio of a firm’s earnings before interest to its total assets, Et/At. The proxies for investment opportunities are a firm’s rate of growth of assets, dAt/At, and its market-to-book ratio, Vt/At. The year t regressions are estimated for all firms with the required data items. The table shows means (across years) of the regression coefficients and t-statistics calculated using the Fama and MacBeth (1973) procedure, further adjusted for potential autocorrelation in the time-series coefficient estimate by adopting the method in Pontiff (1996) using AR(1) error terms. The results are shown in Table 11. Table 11 suggests that firms are less likely to pay dividends the higher the

(6)

(7)

(8)

Fama-Macbeth Regression

0.02 0.05 0.03 0.02 0.03 0.04 0.05 (7.80) (5.37) (7.98) (6.73) (7.57) (12.35) (2.42)  0.017  0.014  0.014  0.014  0.015  0.015  0.007 (3.89) (3.58) (3.57) (3.64) (3.69) (3.33) (2.79) 0.02 0.02 0.03 (8.38) (8.90) (2.58)  0.03  0.02  0.03 (5.88) (3.27) (2.17) 0.00 0.00 0.00 (4.33) (3.22) (1.28) 0.00 0.00 0.00 (1.14) (1.23) (0.18)  0.05  0.04  0.06 (11.95) (7.70) (3.06) 0.00 0.00 0.00 (2.45) (7.08) (1.45) 0.04 0.03 0.04 (24.06) (15.71) (0.93) Y N

Y N

Y Y

Y Y

N N

N N

N Y

0.04 (2.03)  0.005 (2.55) 0.02 (4.52)  0.02 (3.44) 0.00 (0.97) 0.00 (2.45)  0.04 (5.76) 0.00 (3.50) 0.03 (1.81) N Y

0.017

0.017

0.026

0.027

0.019

0.019

0.038

0.039

77,465

77,465

77,465

77,465

77,465

77,465

77,465

77,465

concentration of dividend-averse institutional investors. This is another piece of corroborating evidence that the composition of institutional investors with different tax preferences is related to the firm dividend policy. 5. Conclusion Analyses that treat heterogeneous groups of investors such as institutions homogenously risk conflating tax attributes with other unique aspects of institutional shareholders. The analysis in this paper capitalizes on the heterogeneity within the tax preferences of a type of investor to provide direct evidence of the presence of dividend clienteles. These clientele effects are more pronounced in the restricted sample in which tax preferences are more clearly identified and where those tax preferences are more likely to be pivotal. Such clientele effects are consistent with investors self-sorting to firms with attractive payout policies and with managers altering payout policies in response to shareholder tax-based preferences. These alternative channels are not exclusive, and evidence is provided that both are operative. The evidence in the paper is consistent with cumulative evidence that shows that firms do respond to corporate tax incentives (Auerbach, 2002; Gordon and Hines, 2002; Graham,

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

Table 11 Logit regressions to explain which firms pay dividends. We estimate logit regressions for each year of the 1981–1997 period. The dependent variable is one in year t if a firm pays dividends, zero otherwise. The explanatory variables are the share of dividend-averse institutional shareholders in a firm, profitability (Et/At), the growth rate of assets (dAt/At), the market-to-book ratio (Vt/At), and the percent of NYSE firms with the same or lower market capitalization (NYPt). The year t regressions are estimated for all firms with the required data items. The table shows means (across years) of the regression coefficients and t-Statistics calculated using the Fama and MacBeth (1973) procedure, further adjusted for potential autocorrelation in the time-series coefficient estimate, by adopting the method in Pontiff (1996) using AR(1) error terms. Share of dividend-averse institutional shareholders NYP t Et/At dAt/At Vt/At Number of observations

 0.356 (4.04) 3.47 (16.28) 10.69 (4.80)  1.17 (3.78)  0.61 (3.65) 113,345

2003), and that stock prices incorporate tax factors (Brennan, 1970; McGrattan and Prescott, 2005; Sialm, 2007). Taken together, the evidence in this paper shows that some rational investors (those tax-sensitive institutions) do care about the tax consequences of firm payout policy, and that firms, in turn, do care about their investors’ tax preferences when deciding on their payout policy. Appendix A. The use of the dividend-payout ratio to measure firm payout policy Like the focus of the main results in this paper, earlier papers have also used the dividend yield to study dividend policy. In recent papers, Grinstein and Michaely (2005), Allen and Michaely (2003), Grullon and Michaely (2002), and Blouin and Nondorf (2004) have pointed out that such measures might provide an incomplete picture. If a firm increases dividend-payment and share-repurchase expenditures proportionally, it is difficult to conclude that the dividend policy is more (or less) tilted toward dividends. Rather, such an increase is a demonstration of an overall increase in payouts. To the extent that this critique is valid, to better capture the relative importance of dividends as a form of payout rather than as a change in the overall amount of firm payout, robustness checks are conducted using the dividend-payout ratio as defined below. The main results of the paper survive this robustness check. Following Grullon and Michaely’s (2002) definition, repurchases are measured as expenditures on the purchase of common and preferred stocks (Compustat industrial annual data item #115) minus any reduction in the value (redemption value) of the net number of preferred shares outstanding (Compustat industrial annual data item #56).17 Common dividends are defined as the total dollar 17 Fama and French (2001) propose to measure the repurchase activities by the change in Treasury stocks or the amount of repurchase

83

dividends declared on the common stock during the firm’s applicable fiscal year (Compustat industrial annual data item #21). The ‘‘dividend-payout ratio’’ is defined as the ratio of dividends to the sum of dividends and repurchases. By construction, the measure of the proportion of dividend payments among the total payout is only well-defined for firms that either pay dividends or that repurchase stock during a year. Eliminating firms that do neither during that year reduces by 36,703 the initial sample of 113,345 firmyear observations for which other data are available. The alternative measure of firm dividend policy might have a bias. Firms omitted because they neither pay a dividend nor repurchase stock in a year might be more likely to prefer capital gains to dividends. Compared to paying a dividend or repurchasing, paying nothing leads to an increase in the retained earnings of the firm, thus, the share price will be higher. The practice of retaining earnings in the firm thus creates capital gains rather than dividends on the stock, and researchers might want to treat this as functionally equivalent to a repurchase. The approach described above excludes these observations from the analysis, and thus, might introduce some bias. As an additional robustness check, tests are conducted including firms with zero payout, but nonetheless have a cash-to-asset ratio of 10% or higher, treating them as if they have a dividend/(dividend+repurchase) of zero. These are firms that have a lot of extra cash on hand, but choose not to pay dividends: 14,043 firm-years were reclassified in this way. Using both measures of the dividend-payout ratio, results were obtained that do not qualitatively differ from those reported in the paper.

Appendix B. Definitions of the main variables used in the paper This Appendix describes the main variables used in the paper, along with their Compustat Industrial Annual File code numbers. Cash holdings is defined as the ratio of cash and marketable securities (Compustat Item #1) to total assets (Compustat Item #6). Return on assets is the ratio of operating income (Compustat Item #13) to total assets (Compustat Item #6). Market capitalization is the product of common shares outstanding (Compustat Item #24) and year-end price (Compustat Item #25). Earnings is defined according to the standard accounting literature, for example, Rosett (2001), as Earnings After Interest and Tax (Compustat Item #18). The ratio of capital expenditures to assets is defined as the ratio of capital expenditures (Compustat Item #128) to total assets (Compustat Item #6). Leverage is defined as the ratio of the book value of debt to total assets, where the book value of debt is (footnote continued) minus the amount issued by the firm. As argued by Grullon and Michaely (2002), the measure poses some problems in a fair comparison between dividends and repurchases, since it commingles equity issuance and payments to labor with the true repurchase activity (both equity issuance and payments to labor will affect the Treasury stock). As a robustness check, dividend-payout ratio is also constructed using the alternative measure of the repurchase activities proposed by Fama and French (2001). The results, not reported for brevity, confirm what is reported here.

84

M.A. Desai, L. Jin / Journal of Financial Economics 100 (2011) 68–84

measured as the book value of total assets less the book value of equity (Compustat Item #6—Compustat Item #216). References Allen, F., Bernardo, A., Welch, I., 2000. A theory of dividends based on tax clienteles. Journal of Finance 55, 2499–2536. Allen, F., Michaely, R., 2003. Payout policy. In: Constantinides, G., Harris, M., Stultz, R. (Eds.), Handbook of the Economics of Finance, vol. 1a. Elsevier, North-Holland, pp. 337–429. Amihud, Y., Li, K., 2006. The declining information content of dividend announcements and the effects of institutional holdings. Journal of Financial and Quantitative Analysis 41, 637–660. Auerbach, A., 2002. Taxation and corporate financial policy. In: Auerbach, A., Feldstein, M. (Eds.), Handbook of Public Economics, vol. 3. Elsevier, North-Holland, pp. 1251–1292. Barclay, M., Holderness, C., Sheehan, D., 2006. Dividends and corporate shareholders. Unpublished Working Paper, University of Rochester, Boston College, Pennsylvania State University. Bernheim, D., Wantz, A., 1995. A tax-based test of the dividend signaling hypothesis. American Economic Review 85, 532–551. Blouin, J., Nondorf, M., 2004. Payout policy: a shareholder-level analysis. Unpublished Working Paper, University of Pennsylvania and University of California, Berkeley. Blouin, J., Raedy, J., Shackelford, D., 2003. Capital gains taxes and equity trading: empirical evidence. Journal of Accounting Research 41, 611–652. Brav, A., Graham, J., Harvey, C., Michaely, R., 2005. Payout policy in the 21st century. Journal of Financial Economics 77, 483–527. Brennan, M., 1970. Taxes, market valuation, and corporate financial policy. National Tax Journal 23, 417–429. Brown, J., Liang, N., Weisbenner, S., 2007. Executive financial incentives and payout policy: firm responses to the 2003 dividend tax cut. Journal of Finance 62, 1935–1965. Brunnermeier, M., Nagel, S., 2004. Hedge funds and the technology bubble. Journal of Finance 59, 2013–2040. Chetty, R., Saez, E., 2005. Dividend taxes and corporate behavior: evidence from the 2003 dividend tax cut. Quarterly Journal of Economics 120, 791–833. Dahlquist, M., Robertsson, G., Rydqvist, K., 2006. Direct evidence of dividend tax clienteles. Report No. 51, Swedish Institute for Financial Research. Del Guercio, D., 1996. The distorting effect of the prudent-man laws on institutional equity investments. Journal of Financial Economics 40, 31–62. Desai, M., Gentry, W., 2004. The character and determinants of corporate capital gains. In: Poterba, J. (Ed.), Tax policy and the Economy, vol. 18. MIT Press, Cambridge, MA, pp. 1–36. Dhaliwal, D., Erickson, M., Trezevant, R., 1999. A test of the theory of tax clienteles for dividend policies. National Tax Journal 52, 179–194. Elton, E., Gruber, M., 1970. Marginal stockholder tax rates and the clientele effect. Review of Economics and Statistics 52, 68–74. Fama, E., French, K., 2001. Disappearing dividends: changing firm characteristics or lower propensity to pay? Journal of Financial Economics 60, 3–43.

Fama, E., French, K., 2002. Testing trade-off and pecking order predictions about dividends and debt. Review of Financial Studies 15, 1–33. Fama, E., MacBeth, J., 1973. Risk, return, and equilibrium: empirical tests. Journal of Political Economy 81, 607–636. Gompers, P., Metrick, A., 2001. Institutional investors and equity prices. Quarterly Journal of Economics 116, 229–259. Gordon, R., Hines, J., 2002. International taxation. In: Auerbach, A., Feldstein, M. (Eds.), Handbook of Public Economics, vol. 4. Elsevier, North-Holland, pp. 1935–1995. Graham, J., 2003. Taxes and corporate finance: a review. Review of Financial Studies 16, 1075–1129. Graham, J., Harvey, C., 2001. The theory and practice of corporate finance: evidence from the field. Journal of Financial Economics 60, 187–243. Graham, J., Kumar, A., 2006. Do dividend clienteles exist? Evidence on dividend preferences of retail investors. Journal of Finance 61, 1305–1336. Grinstein, Y., Michaely, R., 2005. Institutional holdings and payout policy. Journal of Finance 60, 1389–1426. Grullon, G., Michaely, R., 2002. Dividends, share repurchases, and the substitution hypothesis. Journal of Finance 57, 1649–1684. Grullon, G., Michaely, R., Swaminathan, B., 2002. Are dividend changes a sign of firm maturity? Journal of Business 75, 387–424. Guay, W., Harford, J., 2000. The cash-flow permanence and information content of dividend increases versus repurchases. Journal of Financial Economics 57, 385–415. Hotchkiss, E., Lawrence, S., 2007. Empirical evidence on the existence of dividend clienteles. Unpublished Working Paper, Boston College. Jagannathan, M., Stephens, C., Weisbach, M., 2000. Financial flexibility and the choice between dividends and stock repurchases. Journal of Financial Economics 57, 355–384. Jain, R., 1999. Essays on agency costs, dividend policy, and corporate ownership structure. Ph.D. Dissertation, UCLA. Jensen, M., 1986. Agency costs of free cash flow, corporate finance, and takeovers. American Economic Review 76, 323–329. Lintner, J., 1956. Distribution of incomes of corporations among dividends, retained earnings, and taxes. American Economic Review 46, 97–113. McGrattan, E., Prescott, E., 2005. Taxes, regulations, and the value of US and UK corporations. Review of Economic Studies 72, 767–796. Michaely, R., Thaler, R., Womack, K., 1995. Price reactions to dividend initiations and omissions: overreaction or drift? Journal of Finance 50, 573–608. Perez-Gonzalez, F., 2003. Large shareholders and dividends: evidence from US fax reforms. Unpublished Working Paper, Columbia University. Pontiff, J., 1996. Costly arbitrage: evidence from closed-end funds. Quarterly Journal of Economics 111, 1135–1151. Poterba, J., 1987. Tax policy and corporate savings. Brookings Papers on Economic Activity 1987, 455–515. Poterba, J., 2004. Taxation and corporate payout policy. American Economic Review 94, 171–175. Rosett, J., 2001. Equity risk and the labor stock: the case of union contracts. Journal of Accounting Research 39, 337–364. Scholz, J., 1992. A direct examination of the dividend clientele hypothesis. Journal of Public Economics 49, 261–285. Sialm, C., 2007. Tax changes and asset pricing. Unpublished Working Paper, University of Texas at Austin. Strickland, D., 1996. Determinants of institutional ownership: implications for dividend clienteles. Unpublished Working Paper, Ohio State University.